I build AI agent infrastructure, developer tools, and trustworthy physical-AI systems — open source, evidence first, and designed to work outside the demo.
My work spans agent tooling, supply-chain security, observability, developer intelligence, and capability-based authorization for robots and edge devices. I care about systems that are useful, inspectable, and quiet in production.
| Project | What it demonstrates |
|---|---|
| EdgePilot LS26 | An evidence-backed Physical AI agent demo for the ListenAI ARCS Mini / LS26, with a zero-dependency simulator. |
| KineGrant Protocol | Open authorization and accountability infrastructure for physical AI: short-lived capabilities, local action gates, replay protection, signed receipts, and post-quantum verification. |
| Developer Intelligence | 196+ read-only tools across 16 developer ecosystems, with no required API keys and rate-limit-friendly caching. |
| Agent Plugin Doctor | End-to-end plugin health checks covering manifests, builds, packaging, installation, runtime compatibility, tests, and CI. |
| Poison Guard | A pre-install supply-chain scanner using AST analysis, deobfuscation, and behavioral heuristics. |
| Agent Replay | A time-travel debugger for replaying, visualizing, and diffing complete agent trajectories. |
- Agent systems: tools, plugins, session lifecycle, observability, evaluation, and developer experience
- Security: supply-chain analysis, sandbox-policy audits, secret redaction, capability-based authorization
- Physical AI: edge-agent prototypes, local action gates, signed receipts, fleet policy, offline verification
- Open-source execution: 70+ public repositories, 15+ npm packages, upstream-ready fixes, reproducible demos, bilingual documentation
Most of my recent work grew out of the DeepSeek Harness ecosystem. The deeper through-line is broader: making autonomous systems easier to build, inspect, secure, and trust.
I'm exploring roles and collaborations with teams building AI agents, developer platforms, security infrastructure, robotics, or edge AI. I work comfortably across TypeScript, Python, JavaScript, Rust, and systems documentation.
If your team values builders who can turn an unclear infrastructure gap into a shipped, testable open-source system, I'd like to hear from you.
signal over noise · evidence over claims · shipping over theater